An AI photo validator for solar installers — feedback before the crew leaves the site.

Freedom Forever installation crews are required to submit proof-of-installation photos before projects can be approved and funded by financing partners. These photos have to meet specific requirements: correct angles, visible components, sufficient clarity, documented milestones.

The problem was structural. Human review wasn't catching every issue before submission. Photos would fail days later in the financing process — after crews had already left the site. When that happened, someone had to go back. A truck roll just to retake photos. Extra labor, project delays, and a direct hit to cash flow across a large operations pipeline.

Roughly 1 in 5 jobs had issues serious enough to create rework or delays. At scale, that's not an edge case — it's an operational cost center.

Before landing on validation, two other fixes were on the table: rebuild the tool that captures photos, or speed up how fast they get submitted. Neither would have caught a bad photo before it left the site — they'd have just made the existing loop faster or shinier.

The core insight was about timing. The failure wasn't happening at submission — it was happening because crews had no way to know a photo was bad until days after the fact. Move the feedback earlier, to the moment the photo is taken, and the problem largely disappears.

The solution is a mobile-first validation tool embedded in the field app. Installers upload a photo, the AI vision model evaluates it against financing requirements, and a result is returned in seconds — with specific, plain-language guidance on what's wrong and how to fix it while they're still on site.

The key design decision in the AI layer was output format. Financing requirements are written in compliance language. The validation result had to be translated into plain, actionable field guidance — not "image does not satisfy milestone documentation standard 4.2" but "panel is partially obstructed — retake from a wider angle."

Two early passes leaned on showing the actual photo next to each flagged item — it felt like the natural approach. Both ran into the same wall: photos dominate the screen at the expense of the fix itself, some checks need more than one photo per requirement, and the most common failure — a missing photo — is the one case a photo-based layout has nothing to show a photo of. The direction that actually held up dropped the photo and led with the fix.

Every design decision was constrained by field conditions. Crews are on ladders and rooftops, often in direct sunlight, with one hand occupied. They need an answer in under three seconds and a clear next step with no ambiguity.

The previous submission flow gave them nothing — upload, submit, wait. The redesign inverts that: the verdict is immediate, specific, and actionable before the crew moves on.

The tool improved the quality of field submissions, reduced preventable financing rejections, and helped accelerate payment by catching photo issues before crews left the job site. Manual review burden on the operations team decreased as submission quality improved upstream.

At scale across a large solar installation pipeline, eliminating unnecessary site revisits and accelerating financing approval has direct cash-flow impact — each avoided truck roll and each faster funding cycle compounds.